Remember the Mannequin Challenge? Yes, they used it too.
Computer vision exploration requires large amounts of data, including real-world imagery. But it can be hard to find, so researchers end up looking for pretty creative places.
The video above shows how researchers used TikTok dances and mannequin challenges to train AI. The quest is for “ground truth,” or real-world examples that can be used to train and evaluate AI based on inferences. The TikTok dataset does this by displaying many movements, clothing styles, backgrounds, and people. This diversity is key to training models that can cope with real-world randomness.
The same thing happens with the Mannequin Challenge. All these people pretending to be still have provided researchers and their models with more real-world data for training than they ever could have hoped for.
You can find this video and the entire library. Vox’s video on YouTube channel.
back to top ↑
